OpenClaw is a self-hosted assistant runtime. One Gateway process on your own hardware holds sessions, tools, events, and channel connections, and the assistant answers in the chat apps you already use: Discord, iMessage, Slack, Teams, Telegram, WhatsApp, Google Chat, Signal, and more than 20 other services. Native apps for macOS, iOS, Android, Windows, and Linux connect to the same Gateway. The same install runs as a personal assistant on a laptop or as a shared team deployment, with configuration as the only difference.
State, memory, and credentials stay on the machine that runs the Gateway. Models and agent harnesses are plugins (Claude, Codex, and local models are examples), so you can swap the model or harness without changing channels or tools, and OpenClaw works with hosted and local model providers alike. By default it contacts nothing except a daily version check, anonymous feature statistics are opt-in, and a single setting disables both.
The Control UI, CLI, and TUI are front ends onto the Gateway. Companion apps and device nodes add voice, Canvas, camera, screen, and device-local actions on supported platforms. Tools, skills, and plugins extend what the assistant can do, and new capabilities are built against a plugin SDK and shared through ClawHub.
Inbound messages are treated as untrusted input: channels that accept direct messages pair unknown senders by default, and an operator approves each pairing. Tools run on the host for the main session unless sandboxing is configured, and guides cover exposing the Gateway remotely and connecting other users. Access to the control plane can be hardened with token auth or a trusted-proxy auth mode.
Features
- Gateway control plane: one local process owns sessions, tools, events, and channel connections, and runs as a personal assistant or a shared team deployment
- Messaging channels: Discord, iMessage, Slack, Teams, Telegram, WhatsApp, Google Chat, Signal, and more than 20 other services
- Native apps and nodes: macOS, iOS, Android, Windows, and Linux apps, with device nodes that add voice, Canvas, camera, screen, and device-local actions
- Swappable models and harnesses: model providers and agent harnesses (Claude, Codex, local models) are plugins exchanged without touching the rest of the setup, hosted or local
- Front ends: a Control UI, a CLI, and a TUI connect to the Gateway, plus slash commands inside chats
- Extensibility: tools, skills, and plugins built on a plugin SDK and distributed through ClawHub
- Local state: memory, credentials, and session state stay on your hardware, with only a daily version check by default and opt-in anonymous statistics
- Sender pairing: DM-capable channels hold unknown senders until an operator approves a pairing code
- Sandboxing: optional sandboxing for tool execution, with guides for exposing the Gateway remotely and onboarding other users
- Control-plane auth: token auth or a trusted-proxy auth mode harden access to the Gateway
